Violent behavior detection based on SVM in the elevator

Abstract

To avoid fighting and violence occurred in the elevator, this paper proposed an abnormal behavior detection method based on SVM to achieve real-time monitoring. Firstly, the corners of the video sequences were detected and the Lucas-Kanade algorithm was used to calculate the optical flow to obtain velocity vector information. Secondly, this algorithm established a feature vector combining the corner kinetic energy with movement characteristics of targets (including change rate of area, change rate of external rectangle length-width ratio, distance between the targets and the angle difference of target movement direction) as the basis of violent behavior detection. Finally, SVM classifier was constructed to identify the violent behavior. The experiment results showed that the method could detect violent behavior in the elevator effectively and the algorithm was with less complex calculation and higher detection rate thus it could alarm real-time.